Breaking through the noise around data and AI in AEC
By Dimitrie Stefanescu · Founder & CEO, Speckle
General notes
It makes zero sense to reverse-engineer, non-deterministically, data that already exists at the source.
Everything in this series examines why that shift, treating source data as the foundation, is necessary to realize AI's full potential in AEC.
Read the first issue →1 of 7 issued
Why AEC firms chasing AI value need to rethink where intelligence actually lives.
Read the first issue →
You already agreed - ISO 19650, every BIM mandate. You just never made it true, and kept handing over PDFs as the real thing. On the industry's oldest broken promise, and why it is suddenly closable.
Purists wait for clean data, so nothing happens. Maximalists pilot AI on top of nothing, so only demos happen. The multiplier is rented and identical for every firm - the data term is the only factor that differs.
Everyone is buying AI; nobody is fixing the data layer underneath. Structured objects, relationships, history, queryability - a prompt only becomes SQL over your model if the substrate exists.
Dirty data is not the problem - finding out it is dirty when it is expensive is. Validation runs on the data, in milliseconds, continuously, every version. Trust is what unlocks automation.
Time-travel every version, prove who changed what and when, quantify the impact. In a dispute or a safety audit that is not a feature - post-Grenfell, it is becoming a legal requirement.
The shift from project-based to continuous data. What becomes possible when a firm's entire design history is queryable, not locked in files - and why the next building should learn from the last three.
At the Source continues on LinkedIn.
A monthly newsletter about data & intelligence in AEC.